Precise motion tracking of piezo-actuated stages via a neural network-based data-driven adaptive predictive controller

Piezo-actuated stages have applications in many areas such as aerospace, semiconductor manufacturing, and biotechnology. However, the inherent hysteresis, creep, and vibration of these stages render it challenging to guarantee tracking precision in positioning control. Although various control strategies based on accurate models of piezo-actuated stages have been developed that show remarkable efficacy, the associated complexity in model development and identification, especially when the system exhibits uncertainties, often presents a hurdle to their practical adoption. In this study, we develop a data-driven control method using an adaptive predictive controller that dynamically obtains an equivalent linear model by estimating the pseudo-gradient of the underlying nonlinear dynamics online using only the input/output measurement data. For controller implementation, a radial basis function neural network is adopted to adjust the controller parameters by leveraging its powerful self-learning adaptation. Tracking control experiments illustrate the effectiveness of the proposed method in comparison with the proportional–integral–derivative controller and classical model-free adaptive predictive controller.